Skip to contentSkip to conversation

2.6.18 w/ GPS time source: worse performance

DISCUSSION Udo van den Heuvel 4 participants 11 messages Oct 17, 2006 – Oct 19, 2006

lore
Share & tools
lore
#0 lore raw

Hello,

Why does a GPS as time source (with ntpd) perform so much worse with 2.6.18?

Udo

#1 lore raw
Quoted 3 lines · Udo van den Heuvel

Hello,

Why does a GPS as time source (with ntpd) perform so much worse with 2.6.18?

Um... you don't give nearly enough information to even begin to know what you're talking about.

Lee

#2 lore raw
Quoted 7 lines · Lee Revell
Quoted 3 lines · Udo van den Heuvel

Hello,

Why does a GPS as time source (with ntpd) perform so much worse with 2.6.18?

Um... you don't give nearly enough information to even begin to know what you're talking about.

?
Offset, error and jitter are higher. That's the performance. I guess that users of a GPS as clock source understand.

#3 lore raw
In reply to Lee Revell
Quoted 6 lines · Lee Revell
Quoted 2 lines · Udo van den Heuvel

Why does a GPS as time source (with ntpd) perform so much worse with 2.6.18?

Um... you don't give nearly enough information to even begin to know what you're talking about.

No one here with a vague idea about the cause for the bad performance? I am sure I not the only one experiencing this.

#4 lore raw
Quoted 10 lines · Udo van den Heuvel
Quoted 6 lines · Lee Revell
Quoted 2 lines · Udo van den Heuvel

Why does a GPS as time source (with ntpd) perform so much worse with 2.6.18?

Um... you don't give nearly enough information to even begin to know what you're talking about.

No one here with a vague idea about the cause for the bad performance? I am sure I not the only one experiencing this.

No, the issue is that a one-sentence bug report is not helpful. You don't give enough information to debug it. Kernel config, steps to reproduce, etc, etc.

Please look at LKML archives for some examples of the type of bug report that does get a response.

Lee

#5 lore raw
Quoted 9 lines · Lee Revell
Quoted 7 lines · Udo van den Heuvel
Quoted 4 lines · Lee Revell
Quoted 1 line · Udo van den Heuvel

Why does a GPS as time source (with ntpd) perform so much worse with 2.6.18?

Um... you don't give nearly enough information to even begin to know what you're talking about.

No one here with a vague idea about the cause for the bad performance? I am sure I not the only one experiencing this.

Quoted 6 lines

No, the issue is that a one-sentence bug report is not helpful. You don't give enough information to debug it. Kernel config, steps to reproduce, etc, etc.

Please look at LKML archives for some examples of the type of bug report that does get a response.

Get the latest kernel.
Compile for your system
Get ntpd.
Configure for your PPS/NMEA source.
Run ntpd.
Watch performance.
Repeat watching of performance.

It is not a crash. Not an oops.
It is stuff that is visible by watching ntpq -pn output, by letting mrtg graph stuff, etc. Watch the offset and jitter collumns. Check /usr/sbin/ntpdc -c kerninfo output. Graph that stuff. Over the monhts you see changes. Some of them you can explain. Some you can't.

I can understand this is not you everyday bug but this *is* an issue.

#6 lore raw
Quoted 4 lines · Udo van den Heuvel

It is stuff that is visible by watching ntpq -pn output, by letting mrtg graph stuff, etc. Watch the offset and jitter collumns. Check /usr/sbin/ntpdc -c kerninfo output. Graph that stuff.

So... you've presumably done that while identifying there is an issue. Please share the results. Have you tried booting back into a 2.6.17 or so and seen offset/jitter improve? etc etc etc.

#7 lore raw

[added John Stultz to cc]

Quoted 9 lines · On Wed, 2006-10-18 at 15:57 -0400, [email protected] wrote:
Quoted 4 lines · Udo van den Heuvel

It is stuff that is visible by watching ntpq -pn output, by letting mrtg graph stuff, etc. Watch the offset and jitter collumns. Check /usr/sbin/ntpdc -c kerninfo output. Graph that stuff.

So... you've presumably done that while identifying there is an issue. Please share the results. Have you tried booting back into a 2.6.17 or so and seen offset/jitter improve? etc etc etc.

Udo,

My crystal ball says this is related to the clocksource changes. But at the very least dmesg and kernel .config is required for debugging.

Also, don't trim CC lists - always use Reply-To-All for LKML threads.

Lee

#8 lore raw
Quoted 12 lines · Lee Revell

[added John Stultz to cc]

Quoted 9 lines · On Wed, 2006-10-18 at 15:57 -0400, [email protected] wrote:
Quoted 4 lines · Udo van den Heuvel

It is stuff that is visible by watching ntpq -pn output, by letting mrtg graph stuff, etc. Watch the offset and jitter collumns. Check /usr/sbin/ntpdc -c kerninfo output. Graph that stuff.

So... you've presumably done that while identifying there is an issue. Please share the results. Have you tried booting back into a 2.6.17 or so and seen offset/jitter improve? etc etc etc.

Udo:
Are you running the linuxpps patches, or is this vanilla 2.6.18 without
any additional patches? Mind sending your dmesg and some "ntpdc -c kerninfo" output? Any of those graphs you mention above would be great as well.

thanks
-john

#9 lore raw
Quoted 12 lines · john stultz
Quoted 10 lines · Lee Revell

[added John Stultz to cc]

Quoted 7 lines · On Wed, 2006-10-18 at 15:57 -0400, [email protected] wrote:
Quoted 3 lines · Udo van den Heuvel

It is stuff that is visible by watching ntpq -pn output, by letting mrtg graph stuff, etc. Watch the offset and jitter collumns. Check /usr/sbin/ntpdc -c kerninfo output. Graph that stuff.

So... you've presumably done that while identifying there is an issue. Please share the results. Have you tried booting back into a 2.6.17 or so and seen offset/jitter improve? etc etc etc.

Quoted 5 lines

Udo:
Are you running the linuxpps patches, or is this vanilla 2.6.18 without
any additional patches? Mind sending your dmesg and some "ntpdc -c kerninfo" output? Any of those graphs you mention above would be great as well.

I am running the latest LinuxPPS patch from Rodolfo Giometti.

# /usr/sbin/ntpdc -c kerninfo
pll offset:           -0.002536 s
pll frequency:        20.610 ppm
maximum error:        0.007567 s
estimated error:      0.00046 s
status:               0001  pll
pll time constant:    2
precision:            1e-06 s
frequency tolerance:  512 ppm
# ntpq -pn
     remote           refid      st t when poll reach   delay   offset
jitter
==============================================================================
 127.127.1.0     .LOCL.          10 l    7   64  377    0.000    0.000
 0.002
*127.127.20.0    .GPS.            0 l    1   16  377    0.000   -0.268
 0.102
-194.109.22.18   193.79.237.14    2 u  257  512  377    5.511    0.951
 1.444
+193.67.79.202   .GPS.            1 u  250  512  377   16.862    0.398
 4.701
+193.79.237.14   .GPS.            1 u  251  512  377   13.306   -0.001
[cut]

I don't have a dmesg for 2.6.18.1 on this machine (did not save it). Will do when there is a next reboot of this VIA CL6000E board.

When the machine was idle offset and jitter would be in the 0.00X range. Now it is more in the 0.XXX range or worse.
Performance for these figures has at least deteriorated a factor of about 10.

The mrtg graphs are of no use since in the month graph the 'good' part just moved off the screen and in the year graph the resolution is too low. :-(

Udo

#10 lore raw
In reply to [email protected]
Quoted 8 lines · [email protected] wrote:
Quoted 3 lines · Udo van den Heuvel

It is stuff that is visible by watching ntpq -pn output, by letting mrtg graph stuff, etc. Watch the offset and jitter collumns. Check /usr/sbin/ntpdc -c kerninfo output. Graph that stuff.

So... you've presumably done that while identifying there is an issue. Please share the results. Have you tried booting back into a 2.6.17 or so and seen offset/jitter improve? etc etc etc.

Did not yet try 2.6.17 again since I only shortly have been looking into this issue.
Other people also see worse performance with 2.6.18 so the need to do the 2.6.17 thing is a bit less.